Detailed Answers to the Questionnaire[BACK TO QUESTIONNAIRE] sent to selected predictors

Group Name Group Number Pure FR Alter method? SS prediction manual intervention homologues domain identification fragment-based approach use other servers lattice-based threading-like potentials relaxation/ optimization/ minimization other information
SAM-T02-human 001 N. Not exclusively FR, although a FR step was always part of the process. N Y. Four neural nets trained on STRIDE, DSSP, STR, ALPHA11 Y. Extensive manual intervention on the harder targets, assembling parts of the model by hand. Inspecting models, modifying cost function, re-optimizing, choosing models. Y. MSA for predicting secondary structure and building an HMM, which is then used for fold recognition and generating fragments. N. Generally started with whole chain, but sometimes broke this up into smaller pieces (not necessarily domains, as they sometimes overlapped) to repeat the fold recognition. Y. Three fragment sources: generic, specific and long-fragment libraries. Y. Used CAFASP output to confirm template selection; sometimes included Robetta models as possible conformations to modify in optimization process. N Y. Not a traditional pair-wise function, but a cost function based mostly on local environmental properties. Does not include hydrogen-bonding term, but does contain pairwise terms for cysteine residues. N -
BAKER 002 N Y. Predictions were ab initio modelled, or modelled in the context of a template, depending on PCons2 score. Y. PSIPred, Jufo, PHD Y. Domain parsing, model selection. Y. Secondary structure prediction, homologues predicted then clustered. Y. Ginzu (in-house) Y. Standard Rosetta library. Y. Pcons to screen targets. N Y. Knowledge-based potential, physically based potential. Y. Monte Carlo minimization; side-chains repacked using MC search through Dunbrack's rotamer library. -
Skolnick-Kolinski 010 N. Uses FR to generate predicted contacts, but these need not come from proteins with the same global fold. N Y. PSIPred N Y. Multiple sequence alignments used to build profiles for threading-based contact prediction. Pair potentials averaged over homologues. N. Domains were only identified if more than one template was found during the threading stage. N N Y Y. Pair, burial, secondary structure prediction. Y. Assembly using replica exchange Monte Carlo. -
ORNL-PROSPECT 012 Y N Y. PROSPECT-SSP (in-house program) Y. Adjustments to minimize the number of long gaps and achieve better structural quality scores, assessed by WhatIf. Y. Profile-profile alignment. Y. Prodom N N N Y. Environment-specific (singleton) and pair contact potentials. N This is the manual version of group 195 (ORNL-PROSPECT server).
S:BAKER-ROBETTA 029 N Y. Predictions were ab initio modelled, or modelled in the context of a template, depending on PCons2 score. Y. PSIPred, SAM-T99, JUFO. N Y. Structures generated for family members then clustered; largest composite clusters submitted as predictions. Y. Ginzu (in-house) Y. Standard Rosetta library. Y. Pcons to find ab initio targets. N Y. Environment and residue pair potentials. N. Robetta uses only a low resolution centroid based representation of sidechains and a straight Monte Carlo search strategy. -
S:Pmodel 040 Y N Y. PSIPred N N N N Y. Consensus of mGenThreader, FFAS, Inbgu, 3D-PSSM, PDB-BLAST, SAM-T99, FUGUE. N N N -
S:Pmodel3 045 Y N Y. PSIPred N N N N Y. Consensus of INBGU-SHGU, FUGUE-2.1, ORFeus. N N N -
SAMUDRALA-NEWFOLD 051 N N Y. PSIPred N. Method is completely automated. N Y. Domains were defined based on the number of secondary structure elements (in-house algorithm). Y. About 30% of moves came from a 3-residue fragments of identical sequence. N N Y. All-atom residue-specific conditional probability scoring function (similar to potential of mean force used by Sippl). Y. ENCAD was used to eliminate any atom-atom overlaps and fix minor steric problems. Same method as used by group 140.
Jones-NewFold 068 N N Y. PSIPred N Y. Energies are averaged over a set of aligned homologous sequences (created using PSI-BLAST). N. Apart from removing domains with obvious sequence similarities. Y. Library of super-secondary motifs and 3-5 residue fragments extracted from 200 high-resolution structures. N N Y. THREADER3 potentials: beta-carbon distance-based potentials of mean force. N -
Friesner 112 N. A combination of fold recognition and simulation methods were used, depending on the target. Y. Different methods were used for alpha-helical and mixed alpha-beta proteins. Y. All secondary structure prediction methods used. Y. Visual inspection of models. Y. Multiple sequence information used to contruct profiles in threading. Y. In some cases identified likely domains from secondary structure prediction. N N N Y. Statistical size-dependent potential. Y. Simulations involved Monte Carlo moves followed by minimization. -
S:I-Sites/Bystroff 132 N Y. If the target has a PSI-BLAST hit, the alignment is used to assign coordinates. Y. HMMSTR. N Y. PSI-BLAST. N. Although sequences were split into overlapping domain-sized segments. Y. I-sites library. N N N Y. Rosetta minimizes the energy during the simulation.
S:PROTINFO-AB 140 N N Y. PSIPred N N Y. Domains were defined based on the number of secondary structure elements (in-house algorithm). Y. About 30% of moves came from a 3-residue fragments of identical sequence. N N Y. All-atom residue-specific conditional probability scoring function (similar to potential of mean force used by Sippl). Y. ENCAD was used to eliminate any atom-atom overlaps and fix minor steric problems. Same method as used by group 051.
FAMS 168 Y N Y. PSIPred N N N N N N N Y. Simulated annealing process using homology modelling software, FAMS. -
ORNL-PROSPECT 195 Y N Y. PROSPECT-SSP (in-house program) N Y. Profile-profile alignment. N N N N Y. Environment-specific (singleton) and pair contact potentials. N This is the automatic server. Group 012 combines the server with manual intervention.
Pushchino 203 N N Y. Mainly PSIPred, but also Jpred, ALB (in-house). Y. Visual inspection of alignments and/or models. Y. Sequence and secondary structure profile used in threading. Y. Identified domains using PSI-BLAST, HMMer and an in-house method. N. Sometimes visual inspection led to good predictions of fragments being merged into a joint model. N N Y. SCF_THREADER uses empirical potentials of short-range interactions N -
3D-PSSM 229 Y N Y. PSIPred N. Fully automated server. Y. Template sequences aligned to PSI-BLAST target sequence profile. N N N N Y. Jones' THREADER solvation potential. N -
Head-Gordon 271 N N Y. PSIPred Y. Visual inspection during global optimization. N N N Y. Used CAFASP summaries to screen out easy FR and CM targets. N N. Physical all atom force field (AMBER) and hydrophobic solvation term derived in-house. Y. Method focusses exclusively on local/global optimization. -
Wolynes-Schulten 294 N N Y. Jpred and also a consensus based on PSIpred, PHD, Jpred, SSPro, and Prof. Y. Visual inspection of top-scoring structures. Y Y. Variety of methods used: a search for exon/intron boundaries, examination of multiple sequence profiles, and results from the PRODOM and CAFASP servers. N Y. CAFASP used to screen targets. N Y. Associative memory terms combined with optimized contact potential and conventional threading potential (which contains pair, hydrogen bond, and profile potentials with a position dependent gap penalty). Y. Energy function minimized using molecular dynamics with simulated annealing. -
Scheraga-Harold 314 N N Y. Jnet, Jpred Y. In most cases, final structures were selected by visual inspection. N Y. In some cases, domain definitions were based on CAFASP results. N Y. CAFASP results were used to detect domains. N N. Used a physics-based energy function (UNRES) with long-range and short-range energy components (including correlation terms). Y. Conformational space annealing; energy minimization. -
Shortle 349 N N Y. PSIPred, in-house method. Y. Extensive visual creation/inspection of models. Y. Used homologoues in secondary structure prediction and threading to find fragments. N. Although sequences were often subdivided, with dividing lines between segments at high turn propensity and low hydrophobicity. Y. Dunbrack's culled PDB set. Y. PSI-BLAST was used to remove comparative modelling targets. Looked at output of some CAFASP secondary structure servers. N Y. Lawrence & Bryant's empirical pair potentials and in-house phi/psi/rotamer propensities and beta-carbon burial propensities N -
Brooks 373 N. Templates were used in cases where alignments were sufficiently good to allow the construction of reasonable (partial) 3D skeletons in agreement with secondary structure predictions. Y. Ab initio was used whenever sufficient template information was not available. Y. Consensus of PSIPred, PHD, SAM-T99, SSpro, as a bias for conformational searching. Y. Visual inspection to select initial templates, manual docking of domains in some cases. N Y. In some cases, domains sampled separately then combined to a single structure. Domain boundaries identified based on alignment with different templates at different parts of the sequence. N Y. CAFASP servers were used to find templates. Y. Ranking and refinement of structures was done using all-atom continuous-space models. N Y. All-atom force-field minimization and molecular dynamics for ranking and refinement. -
Doniach 401 N N Y. PSIPred Y. Picking and ranking submissions. N Y. PSI-BLAST N Y. GenTHREADER used to screen targets. N N N -
TOME 450 Y N Y. Predictions from JPred2 and those included in 3D-PSSM, mGenTHREADER or SAM-T99. Y. Choice of template. Y. Multiple sequence alignments re-threaded onto template to assess structural alignments. Y. Domains sometimes assigned after initial threading runs identified regions corresponding to templates shorted than the target. N Y. Metaserver uses BLAST, 3D-PSSM, mGenTHREADER, FUGUE, SAM-T99 and JPred2. N Y. Pair potential derived from PKB (sidechain/sidechain interactions only). N. Only when the model was built using MODELLER. Main difference between this group and group 464 is choice of template.
ATOME 464 Y N Y. Predictions from JPred2 and those included in 3D-PSSM, mGenTHREADER or SAM-T99. N Y. Multiple sequence alignments re-threaded onto template to assess structural alignments. N N Y. Metaserver uses BLAST, 3D-PSSM, mGenTHREADER, FUGUE, SAM-T99 and JPred2. N Y. Pair potential derived from PKB (sidechain/sidechain interactions only). N. Only when the model was built using MODELLER. Main difference between this group and group 450 is choice of template.
GeneSilico 517 Y N Y. Consensus based on PSIPred, PHD, PROF, APSSP, Jpred, SSPRO. Y. Manual intervention at all stages, including generation of hybrid templates, alignment refinement, and partial rebuilding of backbone and sidechains. Y Y. Fold recognition servers and secondary structure prediction; sequences split manually. Y. Fragments from original fold recognition models, which were positively evaluated using Verify3D. N. Used GeneSilico server. N N N Modified fold recognition approach, selecting best supersecondary structures, optimizing and merging manually.

Consensus or meta servers have answered questions with respect to the overall method, and not component methods.